Richard Carriers Rough Fine-Tuning Argument

Tim Hendrix
January 2016

Introduction

The fine-tuning argument for Gods existence is considered by many atheists and
theists alike to be one of the best arguments for the existence of God. Recently
Dr. Richard Carrier presented what he considerd a definate rebuttel to the
fine-tuning argument in his chapter of The end of christianity (TEC). The
conclusion:1
This entails the Bayesian conclusion that the probability that God
intelligently designed the universe cannot be any higher than 15
percent (and is almost certainly a great deal less than that). That
means no rational person can believe the probability that God intelligently designed the universe is any better than 1 in 6. This means
every rational person must conclude God probably didnt do that.
(RCB)
Tim Hendrix is not my real name, for family reasons I prefer to stay anonymous. Comments or questions may be directed to [email protected]. This is version 2 of the document,
correcting minor typographical and language errors.
1 I have tried to faithfully quote all sources, however I have in places changed typography by
e.g. replacing mathematical symbols and notation with that used in this text for readability.

So not only is the fine-tuning argument supposed to be shown to be false, it is

shown to actually work in reverse and provide evidence against the existence of
God and this must be accepted by every rational person. In this article I will
investigate Dr. Carriers argument in more details and point out where I belive
it fails.
I should say this is in parts a continuation of a discussion with Dr. Carrier on his blog which will henceforth be referred to as RCB (Richard Carrier
Blogs) 2 . the discussion on RCB was in turn a response a follow-up of a set of
articles addressing Dr. Carriers claims by the astrophysicist and writer Luke
Barnes on his blog Letters to nature (LTN)3 and subsequent discussion by Jeffrey Jay Lowder on the secular outpost 4 ; Luke Barnes has written extensively
on astrophysics and fine tuning5 and I strongly encourage a reader to read his
replies to Dr. Carrier and his other articles.
This response will not contain much new over what Luke has already written
so why bother? Firstly, Dr. Carrier still accepts his arguments as being valid
and I believe in the subsequent discussion on RCB he has supported his view
with additional arguments as well as made his position more clear. This article
will respond to these points as they relate to his main arguments and hopefully
provide a concise statement of the central objections Luke and I share. Secondly,
I hope it will be helpful to re-state Dr. Carriers argument and my objection
aimed at a reader who is not familiar with the discussion, has not read TEC
and do not have a background on Bayes theorem. Notice I will solely focus on
Dr. Carriers argument and not discuss my view on the fine-tuning argument
which can be found in the last section for those interested.
Dr. Carriers argument rests upon Bayesian probabilities and a Bayesian
approach to confirmation which will therefore be briefly sketched. The next
sections lays out the fine-tuning argumentin its classical formulation and Dr.
Carriers response. The final part discusses the reasons why I consider Dr. Carriers argument to be fallacious and various attempts by Dr. Carrier to justify
his argument.

Bayesian approach to probability

A Bayesian approach to probabilities considers all statements of interest to be

binary True/False propositions, for instance:
A = The next flip of the coin will be heads
B = God created the universe
C = Saturn has an even number of moons,
Since we do not have perfect knowledge we must reason about these propositions under uncertainty. Our certainty of a proposition A given that another
2 http://freethoughtblogs.com/carrier/archives/9429
3

https://letterstonature.wordpress.com/2013/12/13/probably- not- a- fine- tuned- critique- of- richard- carrier- part- 1/ and https:
//letterstonature.wordpress.com/2013/12/13/probably- not- a- fine- tuned- critique- of- richard- carrier- part- 2/

4 http://www.patheos.com/blogs/secularoutpost/2016/01/17/dr- richard- carriers- rebuttal- to- my-commentary- on- his- exchange- with- dr- luke- barnes-about-the-fine-tuning-argument/
5 http://www.physics.usyd.edu.au/ luke/blog/

proposition B is accepted as true is written as the conditional probability:

p(A|B)
And we say we condition A on B. We write A to indicate the negation of a
proposition A (not A) and AB to indicate the proposition: A and B. In
this way
P (AB|CD)
is read as The probability A is true and B is false given that C is false and D is
true. We need one final ingredient namely the rules of probability theory. For
all propositions A, B and C it holds:
p(AB|C) = p(A|BC)p(B|C)
p(A|C) + p(A|C) = 1
The first rule, the product rule, is saying that the probability of A and B given
C is the same as the probability of B given C times the probability A given BC
and the last rule, the sum rule, says the probability of A and not A must sum
to 1. Keep in mind these rules hold in general and for all propositions. This
document will only use the first rule.

2.1

Reasoning with evidence

Suppose you want to reason about a hypothesis H being true or not true (H) in
light of some evidence E and general background knowledge b. Then we can use
the product rule: p(HE|b) = p(H|b)p(E|Hb) and p(HE|b) = p(E|b)p(H|Eb).
Using these two equalities it follows:
p(H|Eb) =

p(HE|b)
p(E|Hb)p(H|b)
=
p(E|b)
p(E|b)

What we are interested in is p(H|Eb), the probability of H being true in light

of all our evidence and all our background-knowledge. It is more convenient to
use the ratio of H against H: p(H|Eb)/p(H|Eb). If this ratio is larger than 1,
H is more likely on not given the evidence. We can compute:
p(H|Eb)
p(E|Hb) p(H|b)
=

p(H|Eb)
p(E|Hb) p(H|b)

(1)

Notice p(H|b) is the probability our hypothesis is true given no evidence called
the prior probability of H. Usually when we consider an argument the fraction
p(H|b)
is not very relevant since we want to examine how the evidence changes
p(H|b)
this fraction (i.e. makes the hypothesis more or less likely than we a-priori
believed it to be).

2.1.1

Reasoning with two pieces of evidence

Finally, suppose our evidence is composed of two facts, E = e1 e2 . By the

product rule p(E|Hb) = p(e1 e2 |Hb) = p(e2 |e1 Hb)p(e1 |Hb) and so:
p(H|Eb)
p(e2 |e1 Hb) p(e1 |Hb) p(H|b)
=

.
p(H|Eb)
p(e2 |e1 Hb) p(e1 |Hb) p(H|b)

(2)

This is the equation we will most commonly use.

The fine-tuning argument

The fine-tuning argument roughly goes as follows: Our best theories for our
universe contains various constants. If these were different, our universe would
likely be very different and probably not suited for life. For instance if the
expansion of the universe had been slightly larger, stars might not have formed,
and without stars no life. On naturalism, as best as we can tell that the constants
are the way they are is by chance, which is to say it is unlikely they would allow
for life. On the other hand if God exists he would plausibly intend to create a
universe with life and so life-permitting constants and life is plausible on theism.
Hence life and life-permitting constants is evidence towards God.
I wont discuss the physics or theology behind this argument, simply follow
the above intuition and as Dr. Carrier assume the major assumptions behind the
argument are true (for instance that life-permitting values of constants are few,
that God would desire the universe to be full of life, that there is no multiverse,
etc.).
Lets begin by putting the argument in Bayesian terms. First we need to
specify relevant binary variables:
G = God exists
O = Intelligent life, you and me, exists
F = The laws of the universe are compatible with the existence of intelligent life
The negation of G, Atheism, will be written as A = G for simplicity. It is the
plausibility of God we wish to reason about. F requires some qualification. Do
we mean any type of intelligent life or life just as us? I wont deal with that
difficulty here as it will complicate the discussion and simply assume that F
means the constants of nature permits life. Finally we introduce b as6
b = All other background evidence not contained in the above variables.
Our evidence E consists of O and F and our hypothesis is G. Since we know
O, F and b to be true we wish to reason about the fraction:
p(G|OF b)
p(A|OF b)
6 b always enters the same way in the equations and can be ignored. I kept it because Dr.
Carrier prefers it is included and to make the quotes from TEC and RCB more easy to read

If we use the equation for the case of two pieces of evidence we get:
p(G|OF b)
p(O|F Gb) p(F |Gb) p(G|b)
=

p(A|OF b)
p(O|F Ab) p(F |Ab) p(A|b)

(3)

Lets go over these terms one by one. The third term involving p(A|b) and
p(G|b) is the prior probability of the existence (or non-existence) of God before
considering fine-tuning and the existence of intelligent life. When we examine
the fine-tuning argument, we want to investigate how this ratio is affected by
the evidence. The first terms, p(O|F Gb) and p(O|F Ab), is how plausible it is
intelligent life evolved given the universe is fine tuned and God exists or not.
The second terms, p(F |Gb) and p(F |Ab) are how likely it is the universe is
fine-tuned for life given God exists or not.
What you are going to conclude based on the above argument naturally depends on the value one assigns to the six terms. To simplify things, the last
fraction is not very relevant in terms of judging the argument since it is a measure of our a-priori commitment to Gods existence. For simplicity assume we are
roughly uncommitted p(A|b) = p(G|b). The first ratios of probabilities are likely
difficult to estimate, however if we assume that God is the kind of God who got
the universe started with the right laws and then left it along to develop intelligent life we can assume they are roughly equal too: p(O|F Gb) = p(O|F Ab).
This leaves the middle ratio which is the hearth of the fine tuning-argument.
Supposedly, our specific life-permitting laws are unlikely on naturalism (they
could have been so many other ways many presumably unlikely to produce life),
whereas God could have designed them the way they are in order to allow for life
which he desires by assumption. Then p(F |Gb) 1 and p(F |Ab)  1. Suppose
the later are like winning the lottery: p(F |Ab) = 106 then
1
p(G|OF b)
= 1 6 1 106
p(A|OF b)
10
so seemingly this is a strong argument in favor of Gods existence. This argument
can be criticised in a number of ways, however I will simply assume the main
point to get the argument working and focus on Dr. Carriers counter-point.

Dr. Carriers counter-argument

I asked Dr. Carrier twice on RCB if he would clarify his argument by providing
the equation and symbols he intended us to use, however I was told this was in
TEC and his cited material. I have unfortunately not been able to locate his
argument in TEC in a condensed form and his best attempts to describe it is
mixed with analogies and partly scattered over the the many footnotes of TEC.
Below is what appears the most relevant section of TEC and I will leave it to
the reader to determine if it agrees with my summary given below:
Suppose in a thousand years we develop computers capable of simulating the outcome of every possible universe, with every possible arrangement of physical constants, and these simulations tell us which
5

of those universes will produce arrangements that make conscious

observers (as an inevitable undesigned by-product). It follows that
in none of those universes are the conscious observers intelligently
designed (they are merely inevitable by-products), and none of those
universes are intelligently designed (they are all of them constructed
purely at random). Suppose we then see that conscious observers
arise only in one out of every 101,000,000 universes (or whatever ungodly percentage you want, it doesnt matter). Would any of those
conscious observers be right in concluding that their universe was
intelligently designed to produce them? No. Not even one of them
would be. If every single one of them would be wrong to conclude
that, then it necessarily follows that we would be wrong to conclude
that, too (because were looking at exactly the same evidence they
would be, yet we could be in a randomly generated universe just like
them). It simply follows that if we exist and the universe is entirely a
product of random chance (and not NID), then the probability that
we would observe the kind of universe we do is 100 percent expected.
This is not improbable at all, much less too improbable to believe.
(...) The evidence simply always looks exactly the same whether
a universe is finely tuned by chance or by design no matter how
improbable such fine-tuning is by chance. And if the evidence looks
exactly the same on either hypothesis, there is no logical sense in
which we can say the evidence is more likely on either hypothesis.
Think of getting an amazing hand at poker: whether the hand was
rigged or if you just got lucky, the evidence is identical. So the mere
fact that an amazing hand at poker is extremely improbable is not
evidence of cheating. Thus its improbable is simply not a valid argument for design. This seems counterintuitive only because humans
are not well designed for logical or probabilistic reasoning. Those
are skills we have to learn. And they are difficult to learnand even
once learned are still difficult to apply correctly. When we say the
odds are 1 in 101,000,000 that the universe we observe would exist by
chance (or whatever probabilityagain it doesnt matter), we are
erroneously comparing this universe to all other universes that dont
have intelligent life. But we already know we can never be in one of
those universes. This is information we cant ignore. Just like the
people in those simulated universes: they will only ever find themselves in a finely tuned universe whether it was designed or not.
The fact of their universe being finely tuned can never tell them
anything about how it got that way. Once we attend to this correct
logic, we have to compare this universe not to all other universes but
to all other universes we would ever be in. Only if we do that, and
only if there is still a difference between a designed universe and a
chance universe, would we be able to conclude that the universe we
are in was designed (or not) by seeing which of those differences are
observed (or not). Otherwise, we can never tell (TEC)
6

It is my understanding based on careful reading of TEC, RCB and Lukes articles

the argument is as follows: We are still interested in the same ratio as before,
but using the equation:7
p(G|OF b)
p(F |OGb) p(G|Ob)
=

.
p(A|OF b)
p(F |OAb) p(A|Ob)
The argument now goes as follows: If God does not exist and we do exist, the
universe must be fine-tuned for life p(F |OAb) = 1. On the other hand if God
exists and we exists, God could magically cause life to be sustained in a universe
without life-permitting laws, so in general p(F |OGb) =  1 for a number 
(Later Dr. Carrier argues  = 12 but this is not important). Then:
p(G|OF b)
p(G|Ob)
=
.
p(A|OF b)
p(A|Ob)
The argument is simply that p(G|Ob)/p(A|Ob) cant be any larger than p(G|Ob) =
1
1
4 (why this is the case is examined in a moment) and so p(G|Ob)/p(A|Ob) = 3 .
Thus
p(G|OF b)

= .
p(A|OF b)
3
So the fine-tuning of the universe provides evidence against God, and in the
light of fine-tuning Atheism is at least three times more likely than the existence
of God. This is the conclusion which supposedly cannot be rationally denied,
however lets try anyway.

4.1

What went wrong?

Obviously both I and Dr. Carrier follow the rule of probability theory to derive our two respective equations so our disagreements relate to what numbers
we plug into them. Where we disagree is that Dr. Carrier simply assigns a
probability of 41 to p(G|Ob). In doing that, he is not treating O, lifes existence,
like other evidence in a Bayesian framework, but simply asserts God is unlikely
given the most important piece of the evidence namely O. But this is obviously
fallacious! When we consider if life-permitting laws of nature provides evidence
for or against God, we simply cant begin the argument by assuming that given
the most important piece of data, that life exists, atheism is more likely than
theism and proceed from there. It is simply ignoring the main piece of evidence
in the fine tuning argument: that life exists in the universe. This is my objection
to the argument in a nutshell: It is simply ignoring relevant evidence.
A reader can be forgiven for thinking this is a lot of fuzz for a small syntactical difference but in general it makes all the difference. I will illustrate this
with an example:
7 The

reader can check this equation too follows from the product rule

4.1.1

Carrie the Bayesian defence lawyer

Consider the following story: Suppose a woman one day notices blood dripping
from his ceiling. The police is called and in the apartment above the woman they
discover the body of a young girl who went missing the day before. She is lying
in a pool of her own blood with a knife sticking out of her back. The apartment
shows no signs of a break-in. Shortly after the man who lives in the apartment
returns and is quickly arrested.
A jury has to decide if an accused man is guilty or not-guilty G or G and
the evidence at hand is that the bloody body with the knife is Observed at
the accused apparent (O) along with a pool of the victims blood (B). At the
beginning of the trial, the judge has instructed us to be a-priori somewhat
reluctant to find him guilty.
The prosecutors case: The two pieces of evidence is O (body+knife) and B
(pool of blood). The prosecutor then use our above form of Bayes rule:
p(OB|Gb) p(G|b)
p(G|OBb)
=

p(G|OBb)
p(OB|Gb) p(G|b)
Since it is much much more likely to find the bloody body and the victims blood
at the accuseds apparent if he is guilty than if he is innocent, lets say a hundred
times more likely, she argues this is very strong evidence in favor of his guilt:
p(G|OBb)
p(G|b)
= 100
p(G|OBb)
p(G|b)
Which is just confirming our intuition that it is bad for your freedom to have
dead bloodied bodies lying around in your apartment.
Carrie the defence lawyer to the rescure: The accused man has hired
Carrie the defence lawyer who quickly shows the jury the following equation:8
p(B|OGb) p(G|Ob)
p(G|OBb)
=

p(G|OBb)
p(B|OGb) p(G|Ob)
She then argue that given the body is in the appartment, O, we are equally
likely to observe blood regardless of his guilt (p(B|OGb) = p(B|OGb)) because
ofcourse the bloody body is going to leave blood. So, assuming we are a-priori
reluctant to consider him guilty: p(G|Ob) = 13 p(G|Ob) then
p(G|OBb)
1
1
=1 =
3
3
p(G|OBb)
Actually, Carrie argues, notice that the police did not find bloody fingerprints
belonging to the accused (this fact is now considered part of B) in the apartment.
8 Again,

it follows from the rules of probability theory

It stands to reason this is fully explained on the assumption he is innocent

p(B|Gb) = 1, however if he was guilty there would be a chance his bloody
fingerprints would be found: p(B|Gb) =  1. So in fact all the evidence is
evidence against my client being guilty:
p(G|OBb)
1

= =
3
3
p(G|OBb)
She solemnly looks at the jury and declares: This entails the Bayesian conclusion that the probability that my client is guilty given the evidence cannot be
any higher than 15 percent (and is almost certainly a great deal less than that).
That means no rational person can believe the probability that my client is guilty
is any better than 1 in 6. This means every rational person must conclude he
propably didnt do that
Obviously the problem with this tortured logic is Carrie is just asserting
that the most important piece of evidence, the bloody body, does not matter in
terms of the accuseds guilt (p(G|Ob) = 13 p(G|Ob)) and then this assumption is
used to notice another piece of evidence, the blood, becomes irrelevant as well.
My point is the same fallacy is at play with the fine-tuning argument: We assert
the existence of God is unlikely given one important piece of evidence, that life
exists, and then notice another piece of evidence, fine tuning, does not matter
anymore. Ofcourse what Carrie or Dr. Carrier does is a legal move if they can
somehow argue that the correct prior amounts to p(G|Ob) = 14 in either case,
however this argument must include a discussion of how O (the bloody body
or that life exists) relates to our hypothesis G (the man is guilty or that God
exists). This is simply not done in TEC, a main piece of evidence is simply
ignored.
The argument for Dr. Carriers prior probability of p(G|Ob) = 14 is given in
footnote 8:
Note that I have selected this maximum prior[Of Gods existence]
of 0.25 for an additional reason: because no higher prior can be
developed even by pure logic (i.e., assuming no information exists in
b other than bare propositions and logic). Given such zero knowledge
there would be no more than a 5050 chance any self-existent god
exists, and a 5050 chance such a god would be an intelligently
designing god (as opposed to one who was not), and 0.50.5 = 0.25,
which is therefore the maximum possible probability God can have
prior to considering any evidence for or against his existence. (TEC)
This argument is itself more suspicious than a man in a thick trench coat at a
playground in the middle of the Arizona summer, however for our purpose the
prior Carrier develops is as he himself says based on pure logic and therefore
for p(G|b). Most certainly it is not taking into account the central piece of
evidence in the fine-tuning argument, that life exists in our universe O, as would
be required for the probability p(G|Ob).
To see how spurious the argument is, Carrie the lawyer could similarly argue
that she selected a prior of 14 for p(G|Ob) because the prior probability of the
9

client being guilty involves him bringing the girl to his apartment (probability
0.5) and then killing her (probability 0.5)9 and so p(G|Ob) = 12 21 = 14 , but
this is at best a prior for p(G|b) and certainly not the probability of him being
guilty given that a body is found dead in her apartment.
To my mind what Dr. Carrier does is so obviously an illogical move (as the
crime-scene example illustrates) that presumably he can only believe it works
for the particular example where O is special (as it contains our existence)
and therefore belongs together with b where it can be safely ignored as when
the prior probability of 41 is conjoured out of thin air. Dr. Carrier hints at such
a principles existence many times on RCB and in TEC:
All probabilities must be condition on our background knowledge.
And that knowledge already includes the fact that we find ourselves
in a world where life arose and evolved into people. (TEC)
Elsewhere on RCB this is explained to me:
No. O is not in combination with b. O is in b. Just like logic and
mathematics. Because you cannot condition a probability on there
being no observations. Because we are observing. Thus, to make
statements about us you have to condition all probabilities on our
existing. Period. (RCB)
and elsewhere, in direct reply to my question if he had adopted the stance we
should condtion all probabilities on O:
Adopted the stance? Thats not a stance. Thats a mathematical
requirement of Bayes Theorem. You just said you agreed it was. So
why are you now calling it a stance one can choose to adopt?
(RCB)
and many, many other places on RCB. I will call this view, that all probabilities
must be conditioned on O, Carriers Principle (CP). Keep in mind there is no
error in keeping O together with b as in the expression p(G|Ob), we just need to
still take O into account when figuring out this probability. This goes both for
the fine-tuning argument and the case of the bloody body. However Dr. Carrier
does not do that. He simply absorbs O into b and asserts p(G|Ob) = 13 p(G|Ob) as
if the evidence does not exists and presumably believes this is justified based on
the consideration that you cant do anything else because O must go together
with b (Carriers Principle) and so (presumably?) it does not make sense to
derive a proper prior probability for Gods existence, p(G|b). This is at least as
far as I can get to a justification by reading TEC and it corresponds to what
happens formally when the probabilities are computed. It would be very helpful
if the reasoning behind all of this could have been spelled out in a single place,
a point I will return to later.
9 Incidentally, this should also illustrate why this way of assigning priors based on pure
logic is not very rigorous

10

To recap, I think Dr. Carrier believes the above argument is fallacious in

the case of the bloody body, however Dr. Carrier believes that the situation is
different when we are discussing a bloody body, which should be considered as
ordinary evidence, and our existence O. In the following I will try to discuss
the various arguments Dr. Carrier puts forward for his view however first an
argument why CP is wrong:

4.2

The argument for treating lifes existence as other evidence

The argument in favor of treating O, that life exists, as any other piece of
evidence is fairly simple: Since we accept a Bayesian view of confirmation and
O is a variable, O is subject to the rules all variables are subject to and so our
main equation eq. (3) holds. Thats about the extend of it. Its a bit like arguing
if you can add 4 to 5. Why not?

Arguments for CP

lets look at the arguments put forward for CP:

5.1

It cant be observed
(...) I already showed what happens when you push it back a step
[treat O as not being part of background knowledge]. You end up
with Cartesian existence in b. Which entails observers exist. We are
back at 100%. And then I showed what happens when you push it
back even another step, and remove even our knowledge of ourselves
existing from b. You end up making statements about universes
without observers in them. Which can never be observed. (RCB)

Many of Dr. Carriers points revolve around the above intuition: If we dont
condition on O (observers exist), then we are making a statement about a
universe without observers, which can never be observed. There are three points
to this.
Firstly that we do condition on O in the ratio p(G|OF b)/p(A|OF b). Its
just that O does not enter into all the probabilities which it shouldnt10 .
The second point is that not conditioning on O, as in p(G|b), is not assuming there are no observers which would be p(G|Ob).
The third point is the argument is entirely irrelevant: Yes sure we cant
observe a universe without observers in it as a matter of practice, but
after the bloody body is found in the victims apartment we cant observe
a universe where that havent happened either. It is difficult to quite
10 This

too is pointed out many times by Luke Barnes

11

understand why Dr. Carrier disagree with this point since it is very simple
and absolutely central to Bayesian probability theory: Probabilities do not
indicate a factual or causal connection between the variables but a logical
connection in the mind of the person doing the reasoning, thats why as
part of our reasoning we can use (simpler) probabilities which do not
correspond to the full observed state of the word.
There is absolutely nothing frivolous about this: For instance, in logic
many proofs are by contradiction where in order to prove A is true it is
first assumed A is false, then a statement is derived from the assumption
A is false which is known to be false and thereby it can be known A is true.
As part of this argument we have assumed something which is counterfactual (that A is false) but thats entirely asides the matter. Similar, in a
Bayesian computation, you break down the computation into smaller bits
which do not necessarily condition on all available information.

5.2

It cant be observed, take 2

The confusion from above is repeated again and again:

POINT ONE: You cannot get rid of O. Because you have to condition statements on what is true, not on what is false. (...) O is not
in combination with b. O is in b. Just like logic and mathematics.
Because you cannot condition a probability on there being no observations. Because we are observing. Thus, to make statements about
us you have to condition all probabilities on our existing. Period.
(RCB)
Which of course does not amount to a proof but just insisting the principle is
correct. My answer would be a repeat of the above: Not conditioning on O,
i.e. p(G|b), is not conditioning on something which is false, p(G|Ob). But okay,
lets take O together with b as we can do mathematically. But then the prior
for p(G|Ob) has to take O into account. This goes for Dr. Carriers argument
as it goes for Carrie the defence lawyer and both are getting equally invalid
conclusions when they dont do it. A single more example will serve:

5.3

It cant be observed, take 3. Counterfactual edition

You cant have an empty b. For example, logic and mathematics have
to be in b. Or else the theorem is invalidated. Likewise, all knowledge
that cannot be false must go in b. Like that we exist. [this would be
O] Otherwise, you are describing a statement about a universe we
arent in. Thats called a counterfactual. Counterfactuals are not
true of factual states of affairs. If you condition all probabilities on
our not existing, you are describing a state of affairs that doesnt
exist. Its conclusion therefore will not apply to us. (RCB)

12

Once more, to not condition on O as in p(G|b), is not the same as conditioning

on our not existing, p(G|Ob).

This appears to me to be a very elementary point
but instead of repeating it let me give an example: Suppose S is that the sun
shines today and R is that it will rain tomorrow. Then if i say The probability it
the probability
will rain tomorrow (i.e. p(R|b)), this is not the same as p(R|Sb)
it will rain tomorrow given the sun did not shine today. Furthermore it makes
full sense to talk about p(R|b) even if this probability does not mention if the
sun shines today (lets assume it did) and it is therefore counterfactual. In
addition to these comments see the previous points.

5.4

Argument from what is mathematically false

Continuing the above point, I tried many times to point out to Dr. Carrier on
RCB that there is a difference between not conditioning a hypothesis H on O
and conditioning on O. For instance in direct continuation to the previous quote
I wrote: Tim Hendrix: I never said we should condition on life NOT existing
(O). I said we should not condition on O [when computing the prior]. This is
the difference between p(H|b) and p(H|Ob) which are very different things. To
which Dr. Carrier replied:
Not when logical necessity is at hand, when probabilities are 0 and 1.
Because P (O or O) = P (O) + P (O), thus where F = Fine tuning is
observed, P (F |OA or OA) = P (F |OA) + P (F |OA). P (F |OA) = 1
(because if O and A, you can never observe F ). So already we
know P (F |O)11 has to be 0 (unless you dont know how probability
works). But in case you dont know how math works, we can also
know P (F |OA) has to be 0 because if there are no observations,
then Fine tuning is observed is always false (because nothing can
be observed if there are no observations, not even FT). (RCB)
The problem with this argument is that:
P (F |OA or OA)
is most definitely not in general equal to
P (F |OA) + P (F |OA)
which can readily be checked12 . I am not sure if the above argument is just an
afterthought or an important part of Dr. Carriers overall argument for why O
and b have to go together, however it is build on a simple fallacy which could
explain why Dr. Carrier holds to his view.
11 this

is perhaps supposed to be P (F |OA)

instance, if F is that I will find my missing wallet and A is the event it is in my
desk then lets assume p(F |A) = 1 (I am sure I will find it). However if O is some irrelevant
fact like Ozzy Osbournes favorite dish was spaghetti then p(F |AO) = p(F |AO) = 1 but since
(AO or AO) = A (which follows from basic logic) then according to Dr. Carriers computation
we would have that 1 = p(F |A) = P (F |OA or OA) = p(F |AO) + p(F |AO) = 2. This cant
end well.
12 For

13

On a side note, as a rule I would suggest Dr. Carrier checks that his proofs do
not violate basic rules of probability theory when he include them in sentences
like unless you dont know how probability works.
As to his second point, this is a rather silly grammatical argument similar to
the old chestnut where there are laws there has to be a law-giver. In the context
of probability theory observing fine tuning would refer to: We observe constants of nature fall into particular intervals and not: We observe constants
of nature fall into particular intervals and are observed. To give an example,
if I say the probability a particular tree falls in the wood next year is 13 , this is
the probability of the event the tree falls, not the event that the tree falls AND
someone is there to observe it. The argument is just a basic confusion about
language and the meaning of events.

5.5

Unobserved worlds

Some of the arguments are very difficult to convert into logical form. For instance:
POINT THREE: The unlikelihood of a thing happening is not the
probability that it happened once it is observed. Once it is observed,
you have to explain (A) how it happened, not (B) what would be
true if it didnt. It does not matter that there could have been worlds
without observers or how many such worlds there could have been.
Just as it does not matter that there could have been a world with
no rich people in it or how many such worlds there could have been.
If the only way to get rich is x, then when you observe you are rich,
the probability of having gotten rich by x is always 1. No matter how
unlikely x isin other words, no matter how many possible worlds
lack x. You simply arent in that world. So statements about that
world are irrelevant to you. (RCB)
It is very difficult to tell how the above statements should present an argument
for CP. At any rate, sure it matters there could have been worlds without
observers13 . Thats simply because we should favor explanations that easier
explains the evidence than those who have a harder time explaining the evidence.
Its like saying in the example with Carrie: It does not matter there could have
been worlds where there was no body in his apartment. Sure it matters! That
there is a body in his apartment is much easier explained on the assumption the
accused had something to do with it than on the assumption the accused did
not have something to do with it and thats exactly why both common sense
and probability theory indicates it is evidence for his guilt.
Perhaps a reason for the confusion is that Dr. Carrier phrases probabilities
in terms of worlds without observers and not that the probability a universe
(under certain conditions) contain observers is low. This idea, that probabilities
13 Or as I would prefer, since this is what we are talking about, that the probability of
observers existing is low on a particular theory

14

must reflect possibly realized worlds, is I think a consequence of Dr. Carriers

particular interpretation of probabilities as frequencies which is different than
the Bayesian interpretation of probabilities. I will write more about this later.
Finally, if Dr. Carrier tried to put his arguments into formulas I dare say again
he would readily see that the arguments do not support his conclusions.

5.6

An analogy: The killing machine

Dr. Carrier spends much of the space in TEC on analogies. One is the killing
machine:
you are placed in front of a strange machine that shoots thousands
of bullets around the room at random. Was it designed to miss you?
Here answering that question must rely on your background knowledge about strange complex events: are they usually freak accidents
or products of intelligent design? Once we rule out terrestrial design
(as we must in the universe example) and alien design (as we have
stipulated), we are left with no established instances of intelligent design even for complex events, thus our prior probability must reflect
that such instances are exceedingly rare (like instances of rigging
amazing poker hands in a family game). If they ever even happen
at all. (TEC)
Lets check out intuition with an illustration borrowed from Luke Barnes: Imagine you actually go into the room with the killing machine. It spins into action
and shoots bullets all over the room. But you are not hit! Thousands of bullets
fly by your head until the walls are literally riddled with bullet-holes except for
a perfect silhouette where you are standing. Certainly at some point you would
conclude the guy who made the machine rigged it not to hit you?
For fun, lets do a Bayesian analysis of the killing machine. Suppose we have
the variables14
Mn = All n shots fired by the machine missed you
D = The machine was designed to miss you
then we can compute:
P (Mn |Db) p(D|b)
p(D|Mn b)
=

p(D|Mn b)
P (Mn |Db) p(D|b)
We know that if the machine was designed to miss you, then p(Mn |Db) = 1. On
the other hand if the chance of hitting you with each bullet is 1 q, the chance
of missing you in n bullets is q n . So p(Mn |Db) = q n . This gives:
p(D|Mn b)
1 p(D|b)
= n
q p(D|b)
p(D|Mn b)
14 Lukes explanation is more detailed. Without well-defined symbols I really cannot tell who
of us more faithfully translate Dr. Carriers writings to formulas, but we end up concluding
the same thing. See also his writings for other examples.

15

So as more and more shots miss you, you will be more and more inclined to infer
that it is designed to miss you. The example is originally due to Dr. Collins of
human-genomen fame and the relevance to the present discussion is that if the
universe were selecting the laws at random (i.e. as the killing machine shoots
bullets) then the chance of us being alive here would be low, on the other hand
if they were designed to support life, like the machine could be designed to miss
us, the chance of us being here would be high. The point of Dr. Collins is
the killing machine analogy then supports our conclusion, that if we assume
fine-tuning is rare on chance but probable on a designer, then fine-tuning and
our existence is evidence for God.
5.6.1

A killing machine powered by circular logic

To my mind the above conclusion (given the assumptions) is quite obvious,

however lets see how Dr. Carrier turns the conclusion around:
For example, suppose we knew in advance that 1 in 4 such machines
was rigged to miss, and that the chance of their missing by accident
was 1 in 100. Then we would infer design, because in any cohort
of 1,000 victims, on average 250 will survive by design and only 10
will survive by chance, so if you are a survivor your prior odds of
having survived by chance are 10 in 260, or barely 4 percent. From
that prior probability you should conclude design, even when the
evidence (your not getting hit) is the same either way. (TEC)
This is pretty confusing. The first piece of information amounts to p(D|b) = 14
1
and the second, as best as I can tell, to p(DMn |b) = 100
(otherwise it would
not make sense to say that out of 1 000 victims 10 survived by chance). At any
4
. We then obtain:
rate from this we obtain p(D|b) = 43 and so p(Mn |Db) = 300
p(D|Mn b)
300 1
=
= 25
4 3
p(D|Mn b)
Okay so this just confirms our above conclusion with some different numbers.
Doesnt this mean the killing-machine delivers a killing-blow to Dr. Carriers
argument? Dr. Carrier continues:
But suppose you knew in advance that only one in four results was a
product of design, and the others were of chance. Then in any cohort
of a thousand victims you will still know there are on average ten
survivors by chance, but you will also know that for every survivor
there is who survived by design, three more will have survived by
chance, so you will know there can be, on average, only three who
survived by designso if you are a survivor, your odds of having
survived by chance are still three in four or 75 percent. In this case,
you shouldnt conclude designand thats even knowing the odds
of having survived by chance are 1 in 100. So in that case, as in
the scenario just analyzed for our universe, the probability that the
16

machine was designed to miss you is still 25 percent no matter how

unlikely its missing you was. (TEC)
The first piece of information appears to amount to p(DMn |b) = 13 p(DMn |b)
(since in conjunction with the above piece of information a result would be
both that a person had survived as well as whether it was by chance or accident).
Dr. Carrier is then completely correct in concluding that assuming this is the
case:
P (DMn |b)
1
p(D|Mn b)
=
=
3
p(D|Mn b)
P (DMn |b)
What ought to be pointed out is that this is not an argument, it is simply an
assertion that the event the machine wont hit you and it is designed not to
hit you is three times less likely than the event the machine wont hit you and
operates by chance. Thus, after you confused find yourself in the bullet-riddled
room and tries to reason if you survived by chance or design, Dr. Carrier simply
enters the room and informs you this outcome is three times more likely to have
been by chance than by design, end of story. We can then imagine Dr. Carrier
existing the room, the machine turns on again and fires a thousand more rounds
without hitting you, Dr. Carrier enters the room again and informs you this
is still three times more likely by chance than by design and so on and on. we
are simply told what conclusion we should reach and thats it. Similar as with
Carrie, we are simply being told what the probabilities are without dealing with
the evidence at hand.
However there is an even more nasty consequence to this. In the introduction
Dr. Carrier told us the the a-priori chance the machine was rigged not to hit
us was 1 in 4, i.e. p(D|b) = 41 . Then recall the chance of not being hit by the
machine given it is not designed not to hit us is: p(Mn |Db) = p(Mn D|b)/p(D|b)
and so
p(Mn |Db) =

p(Mn D|b)
1 p(Mn D|b)
1 p(Mn D|b)
=
=
= p(Mn |Db)
p(D|b)
3 p(D|b)
3 13 p(D|b)

Please keep in mind this is purely a consequence of Dr. Carriers assumptions

and some symbol manipulation. Allow me to dwell over this result: In order to
disprove the example of the killing machine, Dr. Carrier makes assumptions
about the killing machine which formally amounts to saying it is equally likely
to hit us if it is shooting by chance as it is if it is designed not to hit us.
Dr. Carrier simply ends up assuming design does not matter at all
in terms of how the machine operates and then from this assumption
concludes we cant infer if it was designed or not based on how it
operates.
This is such a perfect example of a circular argument one could use it to
compute the 20th decimal of !
In the case of the killing machine, any reasonable analysis will confirm common sense, that if it misses you after thousands of shots this is very strong
evidence it is designed to miss you. Notice this is true even if it entails so-called
17

counter-factual probabilities such as p(D|b) and p(D|b) where we do not condition on having survived (or not!). Dr. Carriers supposed demonstration that
the killing-machine analogy fails rests on assuming there is no difference in the
chance the machine hits you if it fires at random or if it fires with the intend
of not hitting you; a perfectly circular argument which I think Dr. Carrier only
misses because it is dressed in convoluted sentences and not put in formulas.

5.7

Another analogy: The lottery

Another analogy brought forward is the lottery. It goes as follows:

For example, if 1 in 100 lottery wins are by design, and the remainder
by chance, then the probability that an observed win is by design is
simply 1 in 100, no matter how improbable that win is. Even if the
odds of winning were 1 in 101000000 . For even then, if its still the
case that 99 in 100 wins are the product of chance, then the odds
that this win is the result of chance are not 1 in 101000000 but 99
percent (TEC)
A problem with these illustrations is it is difficult to tell how them being true
or false proves Dr. Carriers point or how they are supposed to translate into
equations, however here goes: First, a lottery where the chance of winning is
101000000 is an incredible difficult lottery to win. The chance of winning the
powerball lottery is by comparison 1 in 292 million, so if we imagine a lottery
which is three times as difficult to win as the powerball lottery, and suppose
the same person wins this lottery every day for 300 years thats still more likely
than the chances of winning the lottery Dr. Carrier have in mind. So suppose
a person did win the powerball lottery day in and day out for years. Surely, we
would at some point suspect cheating right? A Bayesian calculation: Let W be
the event he won the lottery and D be that he won by design (cheating). Then
for the 117th time:
p(D|b)
P (W |Db) p(D|b)
1
p(D|b)
p(D|W b)
=

= 1000000
= 101000000
10
p(D|W b)
P (W |Db) p(D|b)
p(D|b)
p(D|b)
So if we assume the chances of him cheating is a-priori astronomically unlikely,
say 1 to the number of particles in the universe, we would still be exceedingly
confident he did in fact cheat. I am a bit in doubt what the phrase if 1 in
100 lottery wins are by design should refer to in terms of probabilities, but
presumably(?) it means that if a person wins them the chance of that win
being by design is 0.01 or p(D|W b) = 0.01. Dr. Carriers example is now simply
1
99
and p(D|W b) = 100
then (obviously):
saying that suppose p(D|W b) = 100
p(D|W b)
=
p(D|W b)

18

1
100
99
100

1
99

But this is just Carrier asserting what the probabilities are! if we work backwards, Dr. Carrier has assumed the chances of cheating is:
1
p(D|b)
99
= 1000000
101000002
10
p(D|b)

So if we assume the chances of cheating are smaller than the chances of winning
we should not infer cheating when a person wins. Or to translate this into a
statement about the cosmological argument, if we assume the chances of Gods
existence is much much lower than the chances of fine tuning, we should not
assume God exists if fine tuning exists. This is true, but it sure isnt adressing
the fine-tuning argument and it sure isnt compatible with saying the chances
of Gods existence is 14 .

5.8

Dr. Ikeda, Dr. Jeffreys and Dr. Sober has proven the
argument in TEC

A point brought up several times is that the argument Dr. Carrier promotes is
supposedly demonstrated by mathematicians:
Actually this is not my argument. It is the argument of three prominent mathematicians in two independent studies. (RCB)
The three mathematicians are Michael Ikeda, Bill Jefferys15 and Elliott Sober16 .
For instance when I asked Dr. Carrier to clarify elements of his argument on
RCB he responded:
You know, you could just read the papers Im talking about. My
chapter in TEC is detailed and has extensive notes with mathematical notations interpreting the text. The papers by Sober and Ikeda
& Jefferys do the same, from different angles but getting the same
result. You are already repeating much of that material (evidently
unawares) and any questions you have about it are already answered
in those sources. Thats why we wrote them. (RCB)
or in footnote 21 of TEC: All of this is formally proven, and in fully decisive
detail, by Ikeda and Jefferys and by Sober. For brevity let me limit myself to
the paper by Dr. Ikeda and Dr. Jeffreys (IJ). First lets get one thing out of
the way: I agree 100% with everything mathematical written by IJ and my only
reservations might be in the interpretation, more details can be found in two
excellent blog posts by Luke Barnes on LTN where he specifically discusses IJs
argument. However, what IJ argues is not the same as what Dr. Carrier argues.
Let me clarify: what IJ shows as their main result is, in our notation,
p(A|OF b) p(A|Ob)
15 The Anthropic Principle Does Not Support Supernaturalism, http://www.bayesrules.
net/anthropic.html
16 The Anthropic Principle Does Not Support Supernaturalism, http://www.philosophy.
wisc.edu/sober/design\%20argument\%2011\%202004.pd

19

All which is 100% true under the general assumptions this article is written
on. Refreshingly, they even supply a proof and use well-defined symbols which
makes it easy to understand what they have proven and why it is true. Now,
the issue with this result is that it is not saying anything about the ratio we are
really interested in, Gods existence to his Non-existence in light of the evidence
which is what the fine-tuning argument deals with:
p(G|OF b)
.
p(A|OF b)
This is properly best illustrated with an example. Lets return to Carrie the
defence lawyer and recall G is innocence, B is the pool of blood in the accuseds
apartment and O is that we observed a bloody corpse lying on the floor with
a knife sticking out of its back. If we only make the assumption a bloody
body would leave blood (which is no doubt true!) then we get the exact same
inequality as IJ:17
p(G|OBb) p(G|Ob)
But obviously, Carrie the lawyer still has all her work cut out for her to show
the client is innocent because there is still a dead girl with a knife in her back
in the guys apartment.
But if this is true for Carrie, the same must be true for Dr. Carrier who,
after noticing this inequality, still has all his work cut out for him to show this
is evidence God does not exist.
So to summarize, what IJ result describes is the interaction between two
types of evidence, O and F . What Dr. Carrier concludes relate to the probability of his hypothesis being true or false in light of all evidence. These two
things have little in common which the example of Carrie should demonstrate.

The other arguments

The chapter in TEC does not only deal with the cosmological argument but
tries to treat all pieces of evidence normally brought up in natural theology
with essentially the same method as we just saw put to use for the cosmological
argument. Since the difficulties with Dr. Carriers method is perhaps even more
evident for these other pieces of evidence I will briefly discuss them here. First
let me recap the defense strategy of Carrie:
First assume (by pure logic) that given a bloody body in the apartment
O the chance of the client being guilty is p(G|Ob) = 41
Argue details in the evidence is perhaps slightly easier explained if the
client is innocent than if he is guilty (no bloody fingerprints)
Conclude the client is about six times more likely to be innocent than
guilty
17 As

in IJ: p(G|OBb) =

p(B|GOb)p(G|Ob)
p(B|Ob)

1p(G|Ob)
p(B|Ob)

20

p(G|Ob)

To connect this with the above discussion of the cosmological argument, recall
the argument why p(G|Ob) = 41 (the probability of God given we exist) was
based on supposedly pure logic, see the quoted footnote 8 above.
With this three step procedure in mind lets move over the other arguments
of TEC:

6.1

Evolution by natural selection better explains all current life

The evidence at hand is the details about the evolutionary process leading to
current highly-evolved life:
First assume (by pure logic, footnote 8) the chance God exists given
that we (presumably, biological life) exists is no more than p(G|Ob) = 14 .
Argue for about a page certain details about biological life is easier explained by evolution than by theism
Conclude:
Certainly, no rational person can honestly believe the latter
probability is anything above 50 percent. There is simply no
way the odds are 50-50 that a very powerful self-existent being who creates things by design would create current life that
way, exactly the same way evolution would on its own, rather
than any other way thats far more sensible and expected. Yet
that entails the Bayesian conclusion that the probability that
God intelligently designed current life cannot be any higher than
15 percent (and is almost certainly a great deal less than that).
14 That means no rational person can believe the probability
that God intelligently designed current life is any better than
1 in 6. Which means every rational person must conclude God
probably didnt do that. Current life thus does not appear to
be intelligently designed. (TEC)
It is not that I disagree with the conclusion (but I certainly would not put
numbers on it), however is it safe when we consider the evolution of life to
simply start the argument by saying that given that we exist (highly evolved
life), the chance of Gods existence is p(G|Ob) = 14 ?

6.2

Natural biogenesis better explains the origin of life on

earth

The evidence at hand is that life originated on the planet. The argument goes:
First assume (by pure logic, footnote 8) the chance God exists given
that we (presumably, biological life) exists is no more than p(G|Ob) = 14 .

21

 Argue for about a page certain details about the origin of life is easier
explained by naturalism than by theism
Conclude:
This entails the Bayesian conclusion that the probability that
God intelligently designed the origin of life cannot be any higher
than 15 percent (and is almost certainly a great deal less than
that). 17 That means no rational person can believe the probability that God intelligently originated the first life is any better
than 1 in 6. This means every rational person must conclude
God probably didnt do that. The origin of life thus does not
appear to be intelligently designed. (TEC)
Again, I do not think there is evidence God created life at all, however is it safe
when we consider the origin of life to simply start the argument by saying that
given that we exist (life), the chance of Gods existence is p(G|Ob) = 14 ?

6.3

The human mind

This is my favorite: The evidence is the human brain dependent mind. Lets
see how this evidence is treated:
First assume (by pure logic, footnote 8) the chance God exists given that
we (conscious, brain-dependent minds) exists is no more than p(G|Ob) =
1
4.
Argue for about a page certain details about conscious minds and our
biology is easier explained by naturalism than by theism
Conclude:
So the probability wed observe the kind of brain-dependent
mind we do if it was a product of NID must be less than 100
percent, in fact certainly not greater than 50 percent (since we
cant predict from god did it even a 50-50 chance that this is
what hed do, as opposed to all the other options available to
him). Thus we get the same result here weve always gotten:
there cannot possibly be more than a 15 percent chance that
our mind was designed by God. (TEC)
So I agree with the conclusion sans numbers, but I cannot imagine why step 1
should not be seen as simply assuming the conclusion.

Conclusion

As best as I can tell, my disagreement with Dr. Carrier boils down to two
things.
22

The first point: First a simple formal issue. For a hypothesis H, are we
allowed to perform the following computation
p(H|Ob) =

p(O|b)p(H|Ob)
p(O|b)

(4)

Mind this follows from the rules of probability theory, so a negative answer must
invoke some sort of principle whereby O is considered special because it talks
about observers and that somehow prevents us from doing the above. It goes
without saying I have never seen anything that could hint at such a principle
and I simply cant tell where it should be found within the justifications for
a Bayesian account of probabilities I am aware of. Indeed, the entire idea of
such a principle is so alien I tried twice on RCB to simply make Dr. Carrier
acknowledge it explicitly with two main questions which I pointed out were
the most critical to me:
Main Question 1: Do you accept that the above decomposition eq. (4), i.e.
expressing p(H|Ob) in terms of p(H|b) (a probability which is not conditioned on O) is valid? I.e. is that something we are in fact allowed to
do when considering the probability of p(H|Ob)?
Main Question 2: Do you agree eq. (4) follows from the rules of probability
theory? That is, if we are not allowed to use p(H|b), this must necessarily
be because of a philosophical (if you will) reason?
Dr. Carrier unfortunately choose not to acknowledge these two questions but
rather replied to all other elements of my post. I wrote a single short post
asking him if he would provide an answer to these queries whereto he replied
the questions were answered previously (I cant tell where), that the comment
thread was closing but that I could write him on E-mail. I have written an
email with the two above questions and remain hopeful I will eventually get a
reply.
The second point: The second point is that even if we agree we are not
allowed to perform the above decomposition of p(H|Ob) on some grounds, we
should still take O into account when we figure out the probability of H. This
goes for Carrie the Lawyer as it goes for all of us and this plainly does not occur
in footnote 8.
A potential difficulty is that both in TEC and elsewhere Dr. Carrier adopts
his interpretation of finite frequentism when computing priors and interpreting
probabilities. For instance:
Probability measures frequency (whether of things happening or of
things being true). So were really asking how frequently are things
we point to (in all our background knowledge) the product of NID?
Quite obviously, very infrequently indeed. In fact, so far, that frequency is exactly zero. And thats out of a vast number of things
23

weve found exhibiting apparent design. That includes things made

by people (trillions upon trillions of things), things made by animals
(many trillions more), things made by other life-forms, like bacteria
blindly following their genetic computer programs (those programs
may be intelligently designed, but here Im referring only to their
behavior once theyve come to exist, for example, bacteria exhibit
intelligent behavior but are not themselves intelligent; and counting up everything like that, were adding trillions upon trillions of
more examples), and things made by blind physical processes, like
crystallization or the forming of stars from collapsing dust clouds
(processes that, again, may have been intelligently designed but are
not themselves acting intelligently). If we set aside all contested
cases (and take them all out of b and put them in e, so we can
deal with those laterwhich means any alleged design of the universe, the first life, or current life), what do we have left? Not a
single case of NID. And countless billions and trillions of cases of
not NID. It might have been different. We could have had by now
a scientifically confirmed case of NID. Even lots of them. But so far
science has failed to properly verify even one alleged case in the way
that science has established as known facts such things as the laws
of physics or the size and age of the universe. Thus, based on our
background knowledge alone, in other words, before we consider any
as-yet-only-alleged cases of NID, the frequency of NID is practically
infinitesimal (if not, in fact, zero). This means its prior probability
is vanishingly small, approaching zero (TEC)
So on Dr. Carriers interpretation of probabilities, you figure out a prior for
some event by first coming up with a reference class of situations similar to
the one you consider and then use these to come up with your probability. For
instance, you would figure out the probability the man is guilty given the body
is in his apartment by finding all example situations where a man has a dead
body in his apartment and then divide the number of times he is guilty by the
total number of examples. This type of definition fraud with difficulties as it
fully leaves open the definition of reference class and similar. 18
This obviously brings us into problems. In order to use this definition to
obtain the prior probability of God given observers, p(G|Ob), we have to come
up with hypothetical worlds where there are observers and count the number
of those which contains a God. This obviously cant work in practice, and so I
think this is the reason Dr. Carrier falls back to the probability of 41 based on
pure logic or, as in the case above, a reference class of known things. Obviously
these cannot be said to constitute the prior probability in any objective sense
and I therefore do not see why we should rely on the conclusions.
I can say without qualifications that I have never met a statistician who
believed this is a sound procedure for a one-of hypothesis like Gods existence
18 An easily readable account of some of the main arguments is given by Alan Hajek: http:
//joelvelasco.net/teaching/3865/hajek%20-%20mises%20redux%20redux.pdf

24

and in fact it goes directly against the core idea of Bayesian statistics. Page
after page of demonstrations of the fallacies and pitfalls of this view can be
found in the last 50 years of writings on Bayesian epistemology, for instance in
a textbooks such as E.T Jaynes Probability theory (The textbook Dr. Carrier himself recommends in Proving History). I have spend hours trying to
convince Dr. Carrier of this but so far without luck. One would imagine that
he would be inclined to accept a view supported by hundreds of mathematicians and statisticians when he earlier believe three mathematicians supposed
agreement with him to be a sound argument in favor of his view but I digress...
What is even more confusing is that Dr. Carrier insists on calling his collection of methods the only true interpretation of Bayesian probability theory
(this is supposedly proven in Proving history, however I have not been able to
locate the proof anywhere in the book19 ) and any discussion with Dr. Carrier is
unfortunately hampered by his use of technical terms with a different meaning
than what is usually the case.
I believe this is doing Bayesian probability theory a great disservice in the
mind of the casual atheist or christian reader as it surely must give a person
who encounters Bayesian probabilities through Dr. Carriers writings a very
odd view of what it is: Namely a collection of add-hoc devices (the product
rule is always true except when it is not), prior probabilities and reference
classes pulled out of tall hats, arguments that all but assumes the conclusion
and then present their conclusion as something all rational people must except
(or else!) as well as all the inconsistencies and epistemological confusion which
plague finite frequentism; all in a bundle of formulas, analogies, philosophical
considerations, scientific facts, symbols and strongly expressed views of what
the reader can rationally believe in light of quite flimsy arguments.
The bottom line is I hope Dr. Carrier will seriously re-evaluate the conclusions in TEC, both in terms of Bayesian probability theory and (if applicable)
his own view of probabilities. If the argument is found lacking, this is absolutely
nothing to be ashamed of and if on the other hand it is found to be sound, hopefully he will share a clear demonstration of how the conclusion is reached. Both
I and I am sure many others would find this very interesting.

Post-script: My take on the cosmological argument

My take on the cosmological argument consists of three things:

If you are interested in the cosmological argument, listen to a real scientist
like Sean Carroll, Steven Weinberg or Luke Barnes and not a symbolpusher like myself
19 For a discussion of various difficulties in Proving history see https://www.scribd.com/
doc/271358647/Richard-Carrier-Proving-History-Review

25

 Instead of considering two hypothesis A and G, we should consider multiple hypothesis H1 , . . . , Hn consisting of different physical or theistic ideas.
We can compare any two of them:
p(E|Hi )p(Hi )
p(Hi |E)
=
p(Hj |E)
p(E|Hj )p(Hj )
What we then see is that some hypothesis easier explains the evidence
E = OF than others (for instance a multiverse or a Christian God) but
we still cant figure out which is more probable without the priors, p(Hi )
and p(Hj ). This is about as far as Bayes theorem is ever going to bring us
from our armchairs. But did we really expect to solve this problem with
high-school math?
...however under the general assumptions of the fine-tuning argument,
fine-tuning is plausibly weak evidence towards a God or a multiverse over
alternative ideas such as a single universe and the free parameters of nature
(whatever they are) being random (in whatever space they may belong
to). This should not be confused with the statement God or a multiverse
should be accepted as true or highly probable at this point because:
Experience tells us that the only way to tell what theory is true or false
(God, multiverse, etc.) is when it is possible to make fairly specific predictions from the theory which can be tested in experiments, i.e. the
scientific method. This can be put in a Bayesian form if we like or we can
just apply it intuitively. For this reason I do not consider the fine-tuning
argument as proving or providing strong evidence for a christian God (or
a multiverse) at this point.

26